Skip to content
Prev 322874 / 398503 Next

Factors and Multinomial Logistic Regression

On May 2, 2013, at 20:33 , Lorenzo Isella wrote:

            
You still owe us an answer why you thought that this:

Coefficients:
     (Intercept)     science       socst femalefemale
low     1.912288 -0.02356494 -0.03892428   0.81659717
high   -4.057284  0.02292179  0.04300323  -0.03287211

Std. Errors:
     (Intercept)    science      socst femalefemale
low     1.127255 0.02097468 0.01951649    0.3909804
high    1.222937 0.02087182 0.01988933    0.3500151

Residual Deviance: 388.0697 

is at all different from the Stata output. As far as I can tell it is EXACTLY the same!

Apologies for being insistent, but this will come up in Internet searches as "I couldn't make R do what Stata does".
They would appear just to be delta-method based. 

s.e.(f(thetahat)) =~ f'(thetahat) s.e.(thetahat)

in casu f() is exp() and, e.g., looking at the coef. for female in the "low" table:
[1] 0.8847277

(It is a pretty useless quantity. Stata itself doesn't use it for much, either.)
(Intercept)    science      socst femalefemale
low   7.62989469 0.02048619 0.01877141    0.8847053
high  0.02115184 0.02135577 0.02076329    0.3386964